The money went physical. The Next Web reports Jeff Bezos’ Project Prometheus is close to finalising a $10 billion round at a $38 billion valuation, with JPMorgan and BlackRock joining as new investors. The lab launched in November 2025 with $6.2 billion in initial funding and now employs more than 120 people drawn from OpenAI, xAI, Meta, and DeepMind, co-led by CEO Vikram Bajaj, a former Google X scientist and Foresite Labs co-founder. Its stated focus is physical AI for manufacturing, aerospace, robotics, drug discovery, and logistics automation—Bezos’ first operational tech role since leaving Amazon in 2021. Separately, he is raising a $100 billion “manufacturing transformation” fund to acquire and rebuild industrial companies using Prometheus technology. Three of the largest early-stage AI rounds of the week all surfaced within the same news cycle, and two of the three are pointed at physical AI rather than another chat model.
The contrast came from London. Sifted reports Recursive Superintelligence raised $500 million at a $4 billion valuation, led by Google Ventures with participation from Nvidia. The company is four months old, has roughly 20 employees, and was founded by Richard Socher, Salesforce’s former chief scientist, and Tim Rocktaschel, a UCL AI professor and Director at Google DeepMind; the team also includes former OpenAI researchers Josh Tobin, Jeff Clune, and Tim Shi. Recursive’s stated mission is to automate the frontier AI development pipeline so systems improve themselves without human researchers in the loop. Where Prometheus is pointing capital at atoms, Recursive is pointing it at the research process itself—two different theories of where the next order-of-magnitude gain comes from, priced within hours of each other.
The physical-AI thesis is not just a U.S. bet. Tech Startups reports X Square Robot closed a Series B of nearly RMB 2 billion—about $276 million—co-led by Xiaomi and Sequoia China on the same April 20 news cycle. The company builds general-purpose embodied AI systems linking foundation models to real-world robotics, positioning itself at the center of China’s embodied AI capital race. Taken together, the three rounds surface roughly $10.8 billion of prospective AI capital in a single news cycle—with the Bezos round still in talks rather than finalised—and the clear majority of that total is pointed at physical AI or research automation rather than another chat model. The backdrop is that the chat-model scoreboard has converged: Claude Opus 4.7 (57.3), Gemini 3.1 Pro (57.2), and GPT-5.4 (56.8) sit within half a point of each other on Artificial Analysis’s Intelligence Index, a gap the site itself describes as effectively a tie. The capital is reallocating to where the differentiation still is.
Latest from PulseMark
 |
Claude Opus 4.7 vs GPT-5.4 vs Gemini 3.1 Pro Compared
The three frontier chat models effectively tie on the Intelligence Index—but SWE-bench Pro, BrowseComp, and a 35% tokenizer tax on Opus split them sharply by workload. We ran the per-benchmark and per-dollar numbers. |
That’s Tuesday. Three stories, zero fluff.
— The PulseMark Team
Get the Daily Pulse
Sharp analysis on what's actually moving in AI. No hype, no filler, no weekly digest.